@simicvm

Building 🧠 for 🤖

 Tokyo-to, Japan
simicvm.com
Joined June 2018
  • Tweets817
  • Following612
  • Followers317
186Media
🫠
Ben Recht@beenwrektOct 12
OpenAI’s faster Fourier transform is less a mathematical breakthrough and more a malicious exploit of mathematical traditions. argmin.net/p/faster-fourier-…

Faster Fourier Transforms

The OpenAI algorithm for the FFT is less a breakthrough and more a malicious exploit.

argmin.net
67
codex app is the new operating system
2
30
what is Aeon child feature in codex ? @thsottiaux
41
dots + cloud environments is 🤩🤩🤩
13
so now we have @ChatGPT Chat, Work, Codex, Dots, Spaces, Plugins... very confusing.
26
having to use a single dot for every task feels very unnatural. grok bot implementation is the correct one.
1
34
spend enough time working with enterprises and you’ll understand why AI still doesn’t show up in GDP
Alex Kantrowitz
@KantrowitzSep 14
Here's the first episode of my new show with @pitdesi. We talk about the gap between's AI's hype and it's economic impact, the end of days narrative, consumer AI agents, and dads using ChatGPT to plan the best vacations: youtu.be/Kd7ogzVctPw?si=wu1E… w/ @HubSpot

AI Has Transformed Technology But Not the Economy (Yet)

The smartest technology ever built. So why hasn't the economy moved...

youtube.com
2
115
astra light is quite prone to counting mistakes. I've seen similar behavior quite a lot now. on the other hand, correcting itself in the same pass w/o (visible) reasoning trace is impressive.
40
codex is a killer app for iphone duo
75
it also uses temp python scripts to generate html reports
Armin Ronacher ⇌
@mitsuhikoSep 9
Astra is a code golfer.
40
codex still reports duplicate files in the review panel @thsottiaux @jxnlco
37
Astra is AGI. Bug-free web apps, apparently, require ASI.
1
118
there goes my weekend... ✨
3
326
codex has been pulsing the model selector for a while now... is it polling for astra?
3
169
first time seeing this in Codex. Model in question was 5.6 Sol Max in fast mode
60
new numbers after some tweaking are 2K: 728.05 tok/s prefill, 54.47 tok/s decode 16K: 1655.54 tok/s prefill, 50.68 tok/s decode 65K: 1514.95 tok/s prefill, 47.92 tok/s decode
Marko Simic
@simicvmAug 28
glm-5.3-flash no MTP on B300
38
it seems like glm-5.3-flash just can't stop reasoning 🤔
60
glm-5.3-flash no MTP on B300
antirez@antirezAug 28
Enjoy the DwarfStar glm-5.3-flash branch with GLM 5.3 Flash Q2 and Q4 support: single MacBook 128GB or DGX Spark inference, two MacBook RDMA 128GB each Q4 inference in tensor parallel fashion at 37 t/s single generation. ~500 t/s prefill for now, can be higher. ROCm soon.
1
1
1
6
4,203
current Claude and Sol sure love some lanes...
40
qwen3.8 27b and DeepSeek v4 Flash failed to one-shot this task. however, kimi k3 created a beautiful app
Lisan al Gaib
@scaling01Aug 18
literally just tried to let qwen3.8 27b one shot a fluid simulation webapp it thought for 40k tokens, took around 1 hour to generate and it's just black and doesn't work meanwhile Opus 4.5 just oneshots it in a minute and works wonderfully exact same prompt
2
161
Load more